Seven Deadly Sins of Quantitative Investing Using Machine Learning
Unlike traditional econometric techniques and linear models, supervised machine learning (ML) techniques allow for more granular capture of nonlinearity and interactions among input features than a linear model can accomplish. With a linear model, in order to capture higher order interactions the model needs to specifically define interaction terms, which itself is both difficult to predetermine and often impacts the model’s stability. ML algorithms are able to learn such nonlinearities and interactions without a need for predefinition. Such added power comes with potential perils.